Sensitive High Resolution Inverse Detection NMR ... - ACS Publications

A new indirect detection scheme for obtaining 15N/1H shift correlation spectra in .... Journal of the American Chemical Society 2015 137 (3), 1094-110...
0 downloads 0 Views 117KB Size
Published on Web 11/27/2003

Sensitive High Resolution Inverse Detection NMR Spectroscopy of Proteins in the Solid State Eric K. Paulson,† Corey R. Morcombe,† Vadim Gaponenko,‡ Barbara Dancheck,‡ R. Andrew Byrd,‡ and Kurt W. Zilm*,† Contribution from the Department of Chemistry, Yale UniVersity, P.O. Box 208107, New HaVen, Connecticut 06520-8107, and the Structural Biophysics Laboratory, National Cancer Institute, Frederick, Maryland 21702 Received July 16, 2003; E-mail: [email protected]

Abstract: A new indirect detection scheme for obtaining 15N/1H shift correlation spectra in crystalline proteins is described. Excellent water suppression is achieved without the need for pulsed field gradients, and using only a 2-step phase cycle. Careful attention to overall NMR instrument stability was found critical for obtaining the best resolution and sensitivity. Magnetic dilution by deuteration of the protein in combination with highspeed magic angle spinning produces 1H resonances averaging only 0.22 ppm in width, and in some cases lines as narrow as 0.17 ppm are obtained. In application to two different polymorphs of ubiquitin, structure dependent differences in both 15N and 1H amide chemical shifts are observed. In one case, distinct shifts for different molecules in the asymmetric unit are seen, and all differ substantially from solution NMR shifts. A gain of 7 in sensitivity makes the method competitive with solution NMR as long as nanocrystalline samples are available.

Introduction

Whereas the chemistry of proteins is most often studied in fluid solution phase, their structures are more often studied as frozen crystals. The two main characterization tools employed, solution phase NMR and X-ray crystallography, provide complementary windows on their chemistry, structure and dynamics. Solid state NMR (ssNMR) is uniquely positioned to relate crystallographic and fluid phase NMR data and to address important questions in biophysical chemistry. Macromolecular systems too insoluble for liquid state NMR or too disordered for X-ray crystallography can often be most profitably studied using ssNMR methods. One of the primary impediments to wide adoption of ssNMR as a biophysical tool for studying large molecules has been insufficient sensitivity in comparison to solution NMR. This is largely a result of the fact that ssNMR methods almost universally rely on detection of nuclei such as 13C or 15N, which readily provide narrow lines and a high degree of dispersion in chemical shift. In solution NMR, it has long been recognized that indirect detection of the heteronuclei via the sample’s protons1 can provide a significant increase in sensitivity by virtue of the much higher 1H gyromagnetic ratio γ. Even with this recognition and the experimental demonstration of the sensitivity advantage in model systems,2-5 indirect detection of 13C and 15N nuclei via 1H has only very recently been applied to NMR † ‡

Department of Chemistry, Yale University. Structural Biophysics Laboratory, National Cancer Institute.

(1) Maudsley, A. A.; Muller, L.; Ernst, R. R. J. Magn. Reson. 1977, 28, 463469. (2) Ishii, Y.; Tycko, R. J. Magn. Reson. 2000, 142, 199-204. (3) Ishii, Y.; Yesinowski, J. P.; Tycko, R. J. Am. Chem. Soc. 2001, 123, 29212922. 10.1021/ja037315+ CCC: $25.00 © 2003 American Chemical Society

spectroscopy of a small protein.6 In this paper, we present a new ssNMR indirect detection method, which in comparison has superior sensitivity, an 8-fold shorter phase cycle, and does not require pulsed field gradients (pfgs) for water suppression. Using this method, we demonstrate that high quality 15N/1H heteronuclear single quantum coherence (HSQC) spectra can be obtained with 500 nanomoles of nanocrystalline ubiquitin in less than 15 min on a modern 800 MHz NMR instrument. These results are achieved by a combination of careful attention to instrument stability, pulse sequence design, and sample preparation. Average 1H line widths of only 0.22 ppm full width at half-maximum (fwhm) are observed. We also for the first time observe distinct 1H and 15N amide chemical shifts for magnetically inequivalent copies of the same protein in the asymmetric unit, and for different crystal polymorphs of the same protein. Sensitivity in Indirect Detection

There are a number of approaches to calculating the signalto-noise gain in 1H versus X nucleus detection.3,5 We follow Hoult and use the reciprocity principle.7,8 This can be used to directly relate the efficiency in detection to the efficiency in excitation rather than the coil Q. The latter can be difficult to determine, whereas the radio frequency field (RF) amplitude B1w normalized for unit input power7 is readily measured. If (4) Hong, M.; Yamaguchi, S. J. Magn. Reson. 2001, 150, 43-48. (5) Reif, B.; Griffin, R. G. J. Magn. Reson. 2003, 160, 78-83. (6) Chevelkov, V.; van Rossum, B. J.; Castellani, F.; Rehbein, K.; Diehl, A.; Hohwy, M.; Steuernagel, S.; Engelke, F.; Oschkinat, H.; Reif, B. J. Am. Chem. Soc. 2003, 125, 7788-7789. (7) Hoult, D. I.; Richards, R. E. J. Magn. Reson. 1976, 24, 71-85. (8) Hoult, D. I. Concepts Magn. Reson. 2000, 12, 173-187. J. AM. CHEM. SOC. 2003, 125, 15831-15836

9

15831

ARTICLES

Paulson et al.

polarization transfer is used to equalize the 1H and X nuclei differential spin populations, the ratio of signal voltages ξH/ξX following cross polarization for equal numbers of spins then is simply

ξH γ2H(B1w)H ) ξX γ2 (B ) X

(1)

1w X

Assuming t1 and t2 are set to ∼3T2 to provide the same resolution in both experiments, the only additional factors to take into account are the relative line widths ν1/2, and the fractional efficiency f of the additional magnetization transfer required in all indirect detection schemes. For a fixed number of total scans, the indirect-to-direct relative signal-to-noise ratio for a slice in the 2D spectrum is

γ2H(B1w)H

(S/N)H

)f 2 (S/N)X γX (B1w)X

x

(ν1/2)X (ν1/2)H

(2)

This ratio is appropriate for a balanced single coil multiple resonance probe where the 1H and X nucleus RF fields are relatively homogeneous across the sample, and 1H and X nucleus receiver channels have equal noise figures. In our experiments X is 15N, and the 1H operating frequency is 800 MHz. The triple resonance probe9 used has a ratio (B1w)H/(B1w)X ) 0.36. With a short cross polarization (CP) of 150 µs for the magnetization transfer step, the maximum value for f is measured by us to be 0.45. The gain in sensitivity then hinges on the line widths, and the ratio for our experimental situation then in theory is

(S/N)H (S/N)N

) 16.2 ×

x

(ν1/2)N (ν1/2)H

(3)

This last equation emphasizes how critical the relative line widths are in terms of the real achievable sensitivity advantage. Every factor of 2 reduction in 1H line width results in a factor of x2 higher signal-to-noise. However, it is also possible to give the illusion of achieving a sensitivity gain by simply doing a poor job on the X nucleus, for instance by less than optimal 1H decoupling. 1H

Line Narrowing Strategy

Given the importance of achieving the narrowest 1H line widths, we have made an in-depth comparison of several approaches. This work will be reported on in full at a later date and is similar in scope to that recently reported in a study using the R-spectrin SH3 domain as the test sample.6 However, the results are described now in conjunction with the present work as they are quite different in many aspects. These differences most likely stem from the fact that our work has been carried out at a 1H operating frequency of 800 MHz and utilized magic angle spinning (MAS) at 20 kHz, whereas the prior study utilized 500 MHz 1H operation and 10 kHz MAS. 15N and 1H detected 15N/1H correlation spectra have been compared for both 15N enriched and perdeuterated 15N enriched ubiquitin prepared in two different nanocrystalline forms. The latter sample provides fully protonated amides diluted in a (9) Martin, R. W., Paulson, E. K., and Zilm, K. W. ReV. Sci. Instrum. 2003, 74, 1-17. 15832 J. AM. CHEM. SOC.

9

VOL. 125, NO. 51, 2003

perdeuterated matrix by crystallization from normal aqueous media. Proton line narrowing was compared for MAS alone or with frequency switched Lee-Goldburg (FSLG) 10-12 decoupling, both with and without magnetic dilution.13 In brief, we find that 15N detected 15N/1H correlation spectra using FSLG 1H line narrowing have close to the same line widths on a ppm scale in both perprotio and perdeutero protein. This indicates that MAS alone provides sufficient deuteron dipolar decoupling, and that the strength of the 1H-1H dipolar couplings is not the limiting factor in reducing the FSLG line widths. A comparison of the 1H line widths obtained with either FSLG or MAS alone on perdeutero protein finds them equivalent in ppm once the 0.58 FSLG scaling factor is accounted for. This observation would suggest that the line widths are close to being limited by the effective 1H T2, and are scaled according to the fraction of time the spins effectively spend evolving in the transverse plane as expressed by the scaling factor. Application of low level WALTZ-16 15N decoupling during 1H evolution5 improves the 1H line shape in all instances. Unfortunately, X nucleus decoupling during t1 in X nucleus detect FSLG experiments on model compounds is also accompanied by a significant increase in t1 noise. Presumably, this is caused by small amounts of cross polarization transfer14 during decoupling. The best combination of resolution and sensitivity in 1H detect experiments is thus obtained with 15N decoupling and MAS alone on perdeuterated samples. If we take line widths in Hz to directly reflect the spectrum obtained upon Fourier transform of the data,15 the widths we observe for perdeuterated samples are the same when only FSLG is employed as when 15N decoupling and only MAS is applied instead. Because FSLG decoupling scales the 1H chemical shift interaction by a factor of 0.58, the lines must be narrowed with FSLG by more than a factor of 1.7 to achieve an increase in spectral resolution. Said another way, the 1H resolution in HSQC spectra obtained using magnetic dilution are observed on the relevant ppm scale to be better resolved by a factor of ∼1.7 when 15N decoupling is used instead of FSLG line narrowing. This is opposite of the trend observed at 500 MHz with 10 kHz MAS, where the 1H resolution is reported to decrease by a factor of 1.5 on the 1H ppm scale in a similar comparison.6 Magnetic dilution with 15N decoupling then provides the best combination of resolution and signal-to-noise, with the added bonus of requiring less total RF power be applied. Much of this improved resolution can be attributed to the higher field operation here and the twice as fast MAS rate. However, it should also be noted that such results are only obtained if great care is taken to minimize t1 noise by optimizing RF field stability and water suppression. Pulse Sequences

The pulse sequence developed that best addresses these challenges in our hands is diagramed in Figure 1a. 15N (10) Bielecki, A.; Kolbert, A. C.; Levitt, M. H. Chem. Phys. Lett. 1989, 155, 341-346. (11) vanRossum, B. J.; Forster, H.; deGroot, H. J. M. J. Magn. Reson. 1997, 124, 516-519. (12) Vinogradov, E.; Madhu, P. K.; Vega, S. Chem. Phys. Lett. 1999, 314, 443450. (13) McDermott, A. E.; Creuzet, F. J.; Kolbert, A. C.; Griffin, R. G. J. Magn. Reson. 1992, 98, 408-413. (14) Ishii, Y.; Ashida, J.; Terao, T. Chem. Phys. Lett. 1995, 246, 439-445. (15) the reader should be aware that FSLG 1H line widths reported as Hz in ref 6 are defined differently and include a correction for the 0.58 scaling factor.

NMR Spectroscopy of Proteins in the Solid State

ARTICLES

Figure 2. 1H amide only spectrum obtained on ubiquitin nanocrystals prepared as described in the text. The pulse sequence in Figure 1a was used with t1 set to zero. Two scans were acquired. The location where the water resonance would be observed is marked in the figure. This spectrum should be compared to that obtained with pulse field gradients and a 16 step phase cycle in ref 6 with 128 scans.

Figure 1. Pulse sequences used in this work. Solid bars represent π/2 pulses. Unless otherwise noted, pulse phases are set to x. (a) pulse sequence for 1H detect 1H/15N solid-state HSQC spectroscopy. 1H π/2 pulse lengths 2.4 µs. 15N π/2 pulse 4.6 µs. TPPM decoupling performed at 100 kHz. 15N CP transfers used 15N RF field of ∼55 kHz and 1H RF field at ∼75 kHz with a spin rate of ∼20 kHz. t1max is typically 40 ms, and t2 ) 10 ms. The phase φ of the first 15N π/2 pulse is switched between -y and x phases to aquire quadrature data in the indirect dimension. tw typically set to 10 ms. 15N RF level set to ∼2.5 kHz during WALTZ-16 decoupling 1H carrier frequency placed on the water resonance. 15N π/2 pulse prior to final CP and receiver phase switched between y and -y on alternate scans. (b) pulse sequence for 15N detect solid-state HSQC spectroscopy. RF levels and times for π/2 pulses, CP and decoupling as above. t1max typically 10 ms. The 15N frequency can be set to the middle of the amide band for decoupling during t1, and then jumped to the center of the 15N chemical shift range for CP during the φ phase 1H spinlock pulse. This pulse serves to select the desired 1H magnetization component. Typically, the 1H carrier is placed on the water resonance, so quadrature data in t1 is not required. The 1H spinlock and the final CP usually then are set to y phase, but can also be phase stepped together in either a States or TPPI fashion if this is desired. The initial 1H π/2 pulse and the receiver phase are switched between x and -x on alternate scans.

magnetization is created using a fixed level CP of 150 µs. TEDOR transfers have also been used.16,17 The latter give even more stable polarization transfer, but at the cost of 40% less signal. 15N evolution proceeds under TPPM18 or XiX19 decoupling for t1, and the desired component of the 15N magnetization is projected onto the z-axis by a π/2 pulse. Decoupling is continued for a period t1max - t1, followed by a fixed delay for dephasing any remaining transverse 1H magnetization. Transfer of the frequency labeled 15N magnetization back to the 1H is accomplished with a second ∼150 µs CP. The precise length of this pulse is adjusted as part of the water suppression scheme, and the 1H MAS signal finally acquired using WALTZ-16 15N decoupling.5 Only a 2-step phase cycle is used, alternating the phase of the 15N π/2 pulse by 180° on alternate scans. Figure (16) Hing, A. W.; Vega, S.; Schaefer, J. J. Magn. Reson. 1992, 96, 205-209. (17) Schnell, I.; Langer, B.; Sontjens, S. H. M.; van Genderen, M. H. P.; Sijbesma, R. P.; Spiess, H. W. J. Magn. Reson. 2001, 150, 57-70. (18) Bennett, A. E.; Rienstra, C. M.; Auger, M.; Lakshmi, K. V.; Griffin, R. G. J. Chem. Phys. 1995, 103, 6951-6958. (19) Detken, A.; Hardy, E. H.; Ernst, M.; Meier, B. H. Chem. Phys. Lett. 2002, 356, 298-304.

1b also depicts for comparison the pulse sequence used to obtain 15N detected HSQC spectra, constructed following the same design principles. The pulse sequence in Figure 1a is deceptively simple yet effective in achieving suppression of the dominant 1H water resonance. As noted elsewhere, this can be quite challenging. The excellent water suppression demonstrated in the amide only 1H spectrum for ubiquitin shown in Figure 2 was achieved using the pulse sequence in Figure 1a with t1 ) 0 and only 2 scans. Prior to this work, pfgs and a 16-step phase cycle were required to achieve similar levels of water suppression with crystalline protein samples.6 Keeping the proton decoupling on for t1max effectively saturates the 1H magnetization, and using the same time interval for every value of t1 ensures that what water signal does remain, will always be the same. The subsequent 1H spin lock pulse can then have its length, and if necessary its phase, adjusted to null any water magnetization echoing or relaxing prior to the final CP. Presumably incorporation of a pfg pulse during the dephasing delay would further improve performance, although the use of a homogeneity spoiling pulse did not. A second and just as important benefit to adopting the constant time RF scheme is that it greatly enhances overall system stability by keeping the thermal load experienced by the RF power amplifiers and the NMR probe constant throughout the experiment. Care is also taken to make certain that all amplifier noise blanking is done in a constant time manner as well. In addition to providing for superior water suppression, these precautions are essential for achieving sufficient stability in the polarization transfer to eliminate t1 noise and to achieve resolution in the 15N indirect dimension comparable to what is possible in 15N detect mode. Even with this level of experimental care, optimal performance requires careful alignment of the spectrometer on the protein sample itself. Both CP levels must be independently matched, and the decoupling parameters optimized on the experiment as performed. Experimentally, it is observed that 1H decoupling performance as measured by 15N peak height is a sensitive and highly nonlinear function of the decoupling level. Parameters determined in a direct 15N detection experiment are then not sufficiently accurate to be transferred to the 1H detect experiment if the RF power amplifier droops even only 1% over a 40 ms interval. J. AM. CHEM. SOC.

9

VOL. 125, NO. 51, 2003 15833

ARTICLES

Paulson et al.

Experimental Section 15

Perdeuterated N enriched ubiquitin was prepared in E. coli BL21star cells using a plasmid obtained from Prof. Tracy Handel, U. C. Berkeley. The cells were grown in M9 minimal media containing 1 g/liter 15NH Cl and 2 g/liter U-D -glucose made in 98% D O. Isotopically 4 6 2 enriched compounds were obtained from Cambridge Isotope Laboratories, Inc. The protein was purified using reversed phase HPLC and the identity and isotope incorporation was confirmed using electrospray mass spectrometry. Ubiquitin enriched solely in 15N was provided by VLI Research Inc., Malvern, PA. Ubiquitin nanocrystals were prepared using poly(ethylene glycol) (PEG) or perdeuterated 2-methyl-2,4pentanediol (MPD) following protocols detailed elsewhere.20,21 Several milligrams of nanocrystals were packed into 2.5 mm zirconia MAS rotors using precautions that have been described to ensure long-term retention of sample hydration levels.9 All NMR spectroscopy was carried out using a Varian Inc. Inova instrument operating at 799.59 MHz for 1H. The spectrometer is equipped with a high power cavity tuned vacuum tube 1H frequency amplifier, and uses a 1H/13C/15N MAS probe of our own design. The availability of a low noise Varian preamplifier with an actively switched transmit/receive duplexer capable of handling the full 1H decoupling power during 15N evolution was especially critical to the performance of the 1H detect experiment. Pulse parameters are detailed in the figure captions. Power levels required and variable temperature performance of the probe used have been described in detail previously.9 Samples were spun at ∼20 kHz and kept cooled to ∼278 K. Initial spectrometer alignment was performed using a sample of 13C/15N enriched glycine, and subsequent fine-tuning of all parameters performed on the protein samples themselves. Chemical shifts were referenced by relating the 13C chemical shifts of external adamantane22 to the 1H and 15N chemical shift scales.23,24

NMR Results and Discussion

Figure 3 compares 15N and 1H detect 1H/15N HSQC spectra for uniformly 15N and 2D enriched ubiquitin nanocrystals precipitated with MPD and acquired using the NMR pulse sequences shown in Figure 1. The spectrum in Figure 3a is the best 15N detected HSQC acquired to date on this sample preparation. This is compared to an optimized 1H detect spectrum on another sample in Figure 3b. Both data sets were apodized with standard sine-bell resolution enhancement windowing functions to generate the contour maps shown, and indicate the liquidlike resolution that can be achieved. Skyline projections are provided to roughly indicate the level of signalto-noise in the 2D planes. The pair of 2D contour maps display just over 50 separate cross-peaks each, as counted by a computer automated peak pick. The 1H detect experiment used only 2 scans for each of the 192 complex points acquired. With a recycle delay of 1 s these data took only 13 min to acquire after steady state was established. Much more site to site variation in the intensity of the peaks in the 2D map is observed in the 1H detect experiment. In part, this is due to differential cross polarization efficiency, which is magnified in the 1H detect experiment as two CP steps are required. This is also a result of more variation in the widths of the indirectly detected 15N line shapes, which in part is a (20) Martin, R. W.; Zilm, K. W. J. Magn. Reson. 2003, 165, 162-174. (21) Igumenova, T. I. Assignment of Uniformly Carbon-13-Enriched Proteins and Optimization of their Carbon Line Shapes; Columbia University: New York, 2003. (22) Morcombe, C. R.; Zilm, K. W. J. Magn. Reson. 2003, 162, 173-180. (23) Harris, R. K.; Becker, E. D.; De Menezes, S. M. C.; Goodfellow, R.; Granger, P. Pure Appl. Chem. 2001, 73, 1795-1818. (24) Wishart, D. S.; Bigam, C. G.; Yao, J.; Abildgaard, F.; Dyson, H. J.; Oldfield, E.; Markley, J. L.; Sykes, B. D. J. Biomol. NMR 1995, 6, 135-140. 15834 J. AM. CHEM. SOC.

9

VOL. 125, NO. 51, 2003

consequence of the greater difficulty encountered in optimizing the 1H decoupling in the indirect dimension. In general, we find that the resolution in the indirect dimension of either experiment is close to, but never quite as good as, the best that can be achieved in the complementary experiment where that nucleus is directly detected. Drift in either the decoupling power or in the CP condition can easily lead to a doubling of the line widths in the indirect dimension. A higher signal-to-noise 15N detect spectrum was acquired for measurement of line widths and for comparing sensitivity to the 1H detect spectrum in Figure 3b. This experiment used the same sample as in Figure 3b, but 8 times as many scans. Slices along the 1H dimension from both are compared in Figure 4 for the two positions indicated in Figure 3b. We measure these widths using no apodization of the data, and only compare slices with close to the same number of points per Hz. For either direct or indirectly detected experiments, the 1H line widths lie in a range from 0.17 to 0.39 ppm, i.e., from 140 to 310 Hz, with the average close to 0.22 ppm, or 180 Hz. The 1H lines observed in the present work are actually narrower than the 15N lines on the ppm scale, which at ∼35 Hz are ∼0.44 ppm fwhm. The relative merits of the line narrowing provided by 20 kHz MAS, FSLG, or 15N decoupling were most thoroughly investigated on PEG precipitated deuterated nanocrystals. Focusing on a single line at 15N/1H coordinates of 127.3/9.5 ppm, we observe a 1H line width of 280 Hz (0.35 ppm) with MAS alone that narrows to 160 Hz (0.2 ppm) when WALTZ-16 15N decoupling is applied. The same line with FSLG line narrowing but without 15N decoupling is measured to be 160 Hz, or 0.35 ppm when the FSLG scaling factor is taken into account. This latter result agrees well with observations made at 500 MHz, where 1H line widths of 0.35 ppm were also reported using FSLG alone.6 However, in that work when 15N decoupling was applied without FSLG, the line widths increased by a factor of 1.5 in ppm. This is opposite to the result obtained here at 800 MHz and a 20 kHz MAS rate, where the line widths decrease by a factor of 1.7. Overall the resolution in the comparable experiments improves by a factor of 2.6 on moving to the higher field and faster MAS rate conditions. The signal-to-noise advantage of the 1H detect experiment is obvious. In Figure 4 the 15N detect 1H slices were scaled so that the peak heights in Figure 4a and 4b would match. This results in a noise amplitude that is conservatively measured as 2.5 times greater than in the 1H detect experiment. Taking into account the x8 advantage the greater number of scans provides, the improvement in signal-to-noise provided on this sample by 1H detection is a factor of 7. This agrees with eq 3 taking the average 15N line width to be 35 Hz and the average 1H line width as 180 Hz, resulting in a predicted factor of 7.1. Just exactly how much advantage is provided by the 1H detect experiment depends on sample quality and how well the experiments are optimized. The sample used to acquire the 15N detect spectrum in Figure 3a has 15N lines closer to 25 Hz fwhm on average, a consequence of a better sample preparation and the sample responding better to 1H decoupling. Both types of experiments would benefit from better 1H decoupling to achieve more uniformly narrow 15N lines. In general 1H decoupling for 15N seems to be experimentally more demanding, with the optimum parameters occupying a narrower space, than is our experience with 13C. Although there are reasons to expect this

NMR Spectroscopy of Proteins in the Solid State

ARTICLES

Figure 3. Comparison of 1H/15N HSQC spectra taken on MPD ubiquitin nanocrystals. (a) Exceptional example of an 15N detect spectrum. 128 t1 points acquired, using a dwell time of 52 µs in t1. 1H carrier placed on the water resonance. 16 scans per t1 point, 1 s recycle delay. Spectral width in the 15N dimension set to 25 kHz. Unfortunately, this particular sample was inadvertently melted and subsequent preparations have not quite matched the 15N resolution displayed here. The peaks observed for G47, A46, and I36 in a sample of PEG precipitated ubiquitin nanocrystals are shown as red insets. One lower contour was used for the A46 peaks inset as they are low intensity. (b) 1H detect spectrum of MPD ubiquitin nanocrystals using 2 scans per 192 complex t1 points with a dwell time of 208 µs in t1. Pulse sequence parameters for both sets of data as described in the caption to Figure 1.

would be the case,25 a general solution to providing efficient and robust 1H decoupling for 15N MAS ssNMR seems to remain a challenge. In addition to this particular preparation, we have also acquired solid state 1H/15N HSQC spectra for ubiquitin nanocrystals using PEG as the crowding agent.20 This results in another crystal form, with a much larger unit cell and a doubling in the number of lines. This presumably is a consequence of the presence of at least two magnetically inequivalent molecules in the asymmetric unit. The different shifts are then for ubiquitin molecules with subtly different structures. Resolution in the 2D plane is comparable for this sample, but with double the number of resonances the central portion of the 2D map is significantly

more crowded. The peaks for G47, A46, and I36 are still clear outliers, and are easily assigned by analogy to the known solution NMR shifts and the assigned 15N shifts21 for the MPD crystallized form. These peaks for the PEG nanocrystals are depicted as colored insets in Figure 3a to show their relation to the peaks observed in the MPD form. Table 1 collects these shifts, referenced to DSS with a precision of (0.03 ppm.22,23 They vary over 0.5 ppm in the 1H dimension, and up to 1.3 ppm in 15N shift for any single amino acid center, and are quite different from those measured in solution.26 Once X-ray diffraction data is available for these new polymorphs, it will be interesting to see how the shifts correlate with variations in the X-ray determined crystal structures. Such comparisons in

(25) Ernst, M.; Bush, S.; Kolbert, A. C.; Pines, A. J. Chem. Phys. 1996, 105, 3387-3397.

(26) Wang, A. C.; Grzesiek, S.; Tschudin, R.; Lodi, P. J.; Bax, A. J. Biomol. NMR 1995, 5, 376-382. J. AM. CHEM. SOC.

9

VOL. 125, NO. 51, 2003 15835

ARTICLES

Paulson et al.

Figure 4. Assessment of 1H line widths and relative sensitivity of 1H detect and 15N detect HSQC spectra. (a) 1H detect and (b) 15N detect slices showing the 1H spectra obtained at the position corresponding to slice 1 in Figure 3b. (c) 1H detect and (d) 15N detect slices depicting the 1H spectra obtained at the position corresponding to slice 2 in Figure 3b. Although the 15N detect spectra were acquired using 8× as many scans, the noise level is over 2.5× as large. In each case, the data sets Fourier transformed contained 10 ms of evolution for 1H, and 40 ms of evolution for 15N. Since the direct and indirect 1H dwell times were not the same, the data table was zero filled to result in spectra with as close to the same digital resolution as possible. Table 1. Outlying Ubiquitin Solid State and Solution NMR Shifts sample

A46 15N

A46 1H

G47 15N

G47 1H

I36 15N

I36 1H

MPD crystals PEG crystals, molecule 1 PEG crystals, molecule 2 solution phase

132.0 133.1 133.1 132.20

9.4 9.1 8.9 8.81

103.0 101.7 102.4 102.34

9.2 8.2 8.4 8.13

120.0 118.9 118.5 119.88

6.4 6.0 6.1 6.15

Notes: 1H shifts in ppm from DSS as described in ref 22 and 23. 15N shifts referenced according to ref 24. Solution shifts from ref 26. Assignment of the pairs of shifts in the PEG crystals to particular molecules is arbitrary at this time.

combination with chemical shift calculations of the type already successfully applied to studying dynamic protein structures in solution27 will undoubtedly provide a valuable new experimental connection between time averaged solution NMR structures and chemical shifts and X-ray crystal structures via ssNMR. Conclusions

We have reported a new 1H detected 1H/15N HSQC experiment for studying nanocrystalline proteins. Using standard 15N and 2D enrichment to affect magnetic dilution of the 1H nuclei, MAS at 20 kHz is sufficient to produce 1H line widths of only 0.22 ppm. This provides useful dispersion in the 1H dimension in MAS ssNMR of proteins. Spectral resolution and the quality of water suppression are both linked to operating in a manner (27) Pearson, J. G.; Le, H. B.; Sanders, L. K.; Godbout, N.; Havlin, R. H.; Oldfield, E. J. Am. Chem. Soc. 1997, 119, 11 941-11 950.

15836 J. AM. CHEM. SOC.

9

VOL. 125, NO. 51, 2003

that maximally enhances spectrometer stability, an area where further technical innovations would be a welcome improvement. The pulse sequence described herein improves over previous methods by reducing the phase cycle a factor of 8 and eliminating the need for a pfg coil in the MAS probe. An important observation here is the identification of the need for constant power operation in maximizing resolution. The shortening of the experiment to a 2 step cycle makes it a much better building block for incorporation into 3D ssNMR methods for spectral assignment and obtaining tertiary structural constraints. The ability to resolve 1H and 15N chemical shifts for different protein polymorphs and different protein molecules in the same unit cell by ssNMR is an important step toward directly linking NMR chemical shifts to protein structure variations. On the practical side, this highlights the potential of ssNMR as an analytical tool for characterizing distinct forms of a protein in solid-phase preparations, with potential utility in pharmaceutical applications.20 The broader impact of this observation, however, is the recognition that sufficient accuracy and resolution is now attainable by ssNMR to make possible correlation of protein chemical shifts with actual X-ray crystal structures for the same crystal form. Because the solid state is less influenced by large scale molecular dynamics, structure-shift relations determined in this manner are likely to be more direct, and should provide a better foundation on which solution chemical shifts can be related to protein structure and dynamics in the solution phase. In addition to being a significant technical achievement on the path to making ssNMR methods more applicable to problems in structural biology, this work represents a turning point in terms of ssNMR sensitivity required to study large molecules. The work carried out here used a sample occupying a volume of 6.5 microliters and containing approximately 4 milligrams of protein, i.e., only 500 nanomoles. A not atypical solution NMR study might use 300 microliters of a 1 millimolar solution, or about 2/3 the amount of protein. Even with this small sample, the sensitivity of the experiment is such that signal averaging beyond the 2-step phase cycle is not required. From the solidstate NMR spectroscopist’s perspective this is an important milestone. We expect methods of the type presented here in combination with improvements in ssNMR hardware will make possible the adaptation of ssNMR methods to the solution of an ever wider range of problems in biophysical chemistry. Acknowledgment. This work was supported in part by the Wm. M. Keck Foundation and ExxonMobil. C.R.M. gratefully acknowledges the support of a NSERC post-graduate fellowship. The authors gratefully acknowledge A. Brooke and A. Schwartz of Varian Inc. for assistance in acquiring the low noise 1H preamplifier with high power handling capabilities. JA037315+